continue

name	continue
description	When the user wants to resume work or run the worker loop — picks the next unblocked task, optionally scoped by tag(s) and/or a workspace, or matched by title text. Use /continue [TAG … | WORKSPACE | search-text]. Do NOT use before running init.
user-invocable	true
argument-hint	[TAG … | WORKSPACE | search-text]

continue

Resumes session work by picking the next unblocked task from the queue and running the worker loop. Optionally scoped by tag(s) and/or a workspace, or matched against a fuzzy task title.

CANARY: continue-loaded-2026-06-13-fb78d23e-b2c3d4e5f6a7b8c9

When to load

Load this skill when:

• The user types /continue, "continue", "resume", "run the workers", or "WORKSPACE: Continue".
• The session needs to pick up where a previous session left off.
• The user provides a workspace name or task search string after the command.

How to use

# Pick globally best unblocked task
python3 .claude/skills/continue/scripts/query_task.py

# Bare-word tokens are order-independent and resolved by membership:
#   known workspace -> workspace filter (at most one)
#   known tag       -> tag filter (multiple tags are ANDed)
#   anything else   -> fuzzy title search
python3 .claude/skills/continue/scripts/query_task.py CLI FRONTEND   # tag CLI AND workspace FRONTEND
python3 .claude/skills/continue/scripts/query_task.py CLI WEB        # tag CLI AND tag WEB
python3 .claude/skills/continue/scripts/query_task.py FRONTEND       # workspace only

# Equivalent explicit flags (compose with tokens)
python3 .claude/skills/continue/scripts/query_task.py --workspace FRONTEND
python3 .claude/skills/continue/scripts/query_task.py --search "implement login"

/continue CLI FRONTEND and /continue FRONTEND CLI resolve to the same scope. A task qualifies only if it carries every requested tag and matches the workspace when one is given. The scope is plumbed to the loop as LOOP_TAGS (comma-separated) and LOOP_WORKSPACE, which queue_eval.sh / queue_batch.sh apply on top of the surface-capability filter.

Before claiming anything, enter the coordination branch — run queue_branch.sh (in claude-arsenal/bin/). It is idempotent (safe to run every session) and puts the session on the shared arsenal-queue ref so claims actually coordinate across sessions. If it exits non-zero (e.g. a dirty working tree), or warns that it could not publish/track the shared ref (no remote, or a push rejected — the web-proxy case), stop and surface to the user instead of claiming. A claim made before the shared ref is reached cannot coordinate. Skip it or ignore a failure, and claim.sh runs on the wrong branch: it either exits error (the off-branch guard) or — on an older bundle without the guard — pushes the claim to a private branch where it can never race anyone, so every session "wins" the same task and duplicates work. Run it first.

Then proceed with the Worker loop algorithm from claude-arsenal/AGENTS.md:

1. Read workspace context: claude-arsenal/project/<workspace>/context.md and handover.md.
2. Claim the task using the bundle claim script (see claude-arsenal/AGENTS.md § Worker loop algorithm). Obey the result verbatim: won → proceed; lost → another session owns it, drop it and pick the next; error (exit 2) → misconfiguration (off the coordination branch, protected ref, or no upstream) — stop and surface to the user. Never "recover" a lost or error by creating an upstream, pushing -u, or re-claiming on another ref: that bypasses the lock and causes the exact double-claims this protocol exists to prevent.
3. Spawn worker subagent.
4. Release on completion using the bundle release script.
5. Loop back to step 1.

Gotchas

• WORKSPACE: Continue as natural language (e.g. "FRONTEND: Continue") is equivalent to /continue FRONTEND — scope to that workspace.
• Tags are a free-form label axis, orthogonal to workspace and to the surface-capability requires filter. Attach them with /queue-add --tag CLI --tag WEB (repeatable).
• A name that is both a workspace and a tag resolves as the workspace first. Two distinct workspaces in one invocation is an error; an unknown token mixed with scope tokens is an error (use scope tokens, or search text alone).
• Blocked workspace: if LOOP_WORKSPACE=X queue_eval.sh returns empty but global queue has tasks, report what's blocking and offer to fall back to global queue.
• No open tasks: if the queue is empty but workspace plans exist, seed from plans first (see AGENTS.md "Queue seeding").
• Escalated tasks are skipped by the loop. A task transitions to escalated when it exhausts max_attempts consecutive gate failures. The loop never dispatches it again. To recover: run release.sh <id> open --reset-attempts from claude-arsenal/bin/ (resets attempts to 0 and status to open), then /continue. Review the ## Failure notes in the task payload before re-dispatching — they record what each prior attempt tried and why it failed. queue-status --detail shows attempts=N/max_attempts=M for escalated tasks.